Anthropic Mythos shaping up as nothingburger

Anthropic's highly anticipated AI system, Mythos, has been met with widespread disappointment, with early reviews and community discussions on Reddit labeling it a 'nothingburger.' The system, which was expected to showcase significant advancements in reasoning, coding, and safety—areas where Anthropic has previously excelled—has instead delivered only marginal improvements over existing models like Claude Opus 4.5. Benchmarks reveal a mere 5-10% lift in performance on standard tests such as MATH and HumanEval, while inference costs remain high at roughly $0.15 per 1K tokens. This underwhelming showing has led many to question whether the hype around Mythos was overblown or if the company's research pipeline has hit a plateau.

Critics point to a lack of new architectural innovations or training techniques, suggesting Mythos may simply be a fine-tuned version of an older model. The system's safety features, once a key selling point for Anthropic, have also been criticized for being overly restrictive, limiting practical use cases. With competitors like OpenAI and Google releasing more capable models, Mythos's failure to impress could erode Anthropic's standing in the AI race. The company has yet to release an official statement, but the backlash highlights the growing pressure on AI labs to consistently deliver breakthroughs in a rapidly evolving market.